59th EASTERN SNOW CONFERENCE
Stowe, Vermont USA 2002
Remote Sensing of Snow Cover for Climate Monitoring
in the Canadian Subarctic: A Comparison Between
SMMR–SSM/I and NOAA–AVHRR Sensors
FRÉDÉRIQUE C. PIVOT,1 CLAUDE R. DUGUAY,1,2, ROSS D. BROWN,3
BERTRAND DUCHIRON,4 AND CLAUDE KERGOMARD,5
ABSTRACT:
By considering the advantages and disadvantages of each remote sensing observation method,
the most favourable solution for studying snow cover currently remains the synergistic use of
multi-sensor satellite data. Using Principal Component Analysis (PCA) applied to SMMR and
SSM/I time series over a 23-year period (1978–2001), we were able to evaluate the sensitivity of
passive microwave sensor data in identifying the seasonal cycles and the interannual variability of
snow cover conditions in the Canadian Subarctic (northern Manitoba). We also demonstrated that
the Nimbus-7 SMMR Pathfinder Daily EASE-Grid Brightness Temperatures are biased, especially
in vertical polarisation. Dates characterising the beginning of snow accumulation on the ground
and snow cover disappearance were obtained from the PCA and then compared to both snow
cover information derived from optical satellite imagery (NOAA/NESDIS weekly snow charts)
and conventional snow cover measurements recorded at the Churchill and Gillam airport weather
stations. It appears that passive microwave radiometers meet some difficulties in estimating
accurately the date of the first snow on the ground, especially under warm and dry winter
conditions. Indeed, the difference can sometime exceed 4 weeks (e.g. winters 92–93 and 98–99)
between passive microwave estimates and in situ observations. Consequently, the results clearly
indicate the need to improve snow cover duration estimates derived from passive microwave data
by adding information from optical sensors. However, SMMR and SSM/I data efficiently
compensate for the deficiencies of optical imagery, especially when cloud cover remains over a
study area for long periods due to the presence of large atmospheric disturbances. In addition,
passive microwave data also allow to monitor intra- and inter-annual variations in snow depth.
Keywords: snow cover, remote sensing, multi-sensor approach, passive microwave data, SSM/I,
SMMR, NOAA-NESDIS weekly snow charts, daily data and pentads.
1
LTMEF, Département de Géographie et Centre d'Études Nordiques, Université Laval, SainteFoy (QC), G1K 7P4, Canada.
2
Geophysical Institute, Department of Geology and Geophysics, University of Alaska
Fairbanks, 900 Yukon Drive, P.O. Box 755780, Fairbanks, AK, 99775-5780, U.S.A.
3
Meteorological Service of Canada, 2121 Trans Canada Highway, Dorval (QC), H9P 1J3,
Canada
4
Laboratoire d'Océanographie Dynamique et de Climatologie, UMR 7617, NRS/IRD/Université
Pierre et Marie Curie, Institut Pierre Simon Laplace, 75252, PARIS Cedex 05, France.
5
Géographie des Milieux Anthropisés, CNRS FRE 2170, Bâtiment de Géographie, Université
des Sciences et Technologies de Lille, 59 655, Villeneuve d’Ascq Cedex, France.
15
59th EASTERN SNOW CONFERENCE
Stowe, Vermont USA 2002
INTRODUCTION
For the climatologists and the meteorologists studying atmospheric and climatic variability on a
global scale, the monitoring of the snow cover component undeniably remains a challenge.
Through its inherent physical properties, snow cover significantly influences the evolution of
weather on a daily basis as well as climate variability on a much longer time scale. First snow
modifies the radiative balance and energy exchanges between the surface and the atmosphere, and
thus indirectly acting on the atmospheric dynamics. In addition, this component of the cryosphere
is extremely sensitive to climate fluctuations. Consequently, the accurate observation of the spatial
and temporal variability of snow cover could contribute, in a foreseeable future, to the monitoring
of Global Change.
To study the effects of the extent and variability of seasonal snow cover on the climate system
and for data assimilation or for the validation of GCMs, climatologists need to acquire snow cover
information both on a fine temporal scale and large areas. Compared to the conventional snow
measurements sporadically taken in situ, satellite remote sensing data are particularly well-adapted
to the monitoring of snow-covered surfaces over continuous space-time scales. Nowadays, a wide
range of instruments is available for measuring and observing snow cover; especially, a variety of
spaceborne sensors with various spectral, spatial and temporal resolutions that purposely meets the
needs both required by climatologists and hydrologists. However, each type of sensor is has some
specific technical constraints and limits. Visible-band and near-infrared (NIR) imagery, for
example, only offers the possibility to map the evolution and extent of snow cover. Furthermore,
the accuracy of the snow extent maps inferred from shortwave data remains largely dependent of
cloud cover and illumination conditions, especially in the higher latitudes which are strongly
affected by darkness throughout the long polar nights. The use of thermal infrared images allows
surface temperature monitoring and also complements the snow cover extent mapping. At these
wavelengths, nocturnal observations can be collected Nevertheless, the thermal infrared signal
remains sensitive to the presence of cloud cover. As they are completely independent of
illumination conditions and nearly unaffected by cloudiness, a great deal is expected from
spaceborne passive and active microwave data. Indeed, most clouds are transparent in this portion
of the electromagnetic spectrum (Hall and Martinec, 1985). In homogeneous areas, snow cover
parameters such as snow cover extent and snow water equivalent can be inferred on a daily basis
using passive microwave data. The all-weather investigation of the spatio-temporal variability of
snow packs thus represents the major application of passive microwave radiometry (Gloersen et
al., 1984). However, because of their coarse spatial resolution, the use of current passive
microwave sensors is restricted to the climatological and hydrological studies conducted at
regional to global scales. In contrast, the fine spatial resolution of Synthetic Aperture Radars
(SAR) makes active microwave data a viable tool for studying snow cover, on a fine scale, in the
context of hydrological investigations. However, the complex interactions between the radar
signal and snow pack properties strongly limits the processing and interpretation of radar images.
In addition, the 5.3 GHz frequency (C-band) at which current SAR sensors operate is not
optimised for dry snow cover studies, especially for a quantitative determination of snow cover
parameters such as snow water equivalent (Rango, 1993).
Presently, because of the advantages and disadvantages inherent to each data type employed to
monitor snow cover from space, the most favourable solution probably consists of using jointly
multi-sensor satellite data (visible-band and near-infrared, thermal infrared and microwave). This
strategy is increasingly highlighted in a growing number of publications. Jin and Zhang (1999),
for example, simultaneously used passive and active microwave data to significantly expand the
temporal monitoring of the complex snow covers encountered in Siberia and Greenland. By
merging the data issued by the SSM/I passive microwave radiometer with the information inferred
from the OLS thermal infrared sensor mounted on the same spacecraft, Standley and Barret (1999)
attempted to reduce the confusion between the slightly snow-covered surfaces and clouds
containing a strong amount of microwave scatterers. Others have combined traditional snow cover
information with optical (NOAA–AVHRR imagery) and radar (ERS-2) data in order to
16
periodically update hydrological models inputs, thus substantially improving streamflow
forecasting during the snowmelt period (Koskinen et al., 1999).
In this paper, we present preliminary results of a study related to the monitoring of climate
variability in the Canadian Subarctic (Northern Manitoba) which was recently undertaken in order to
assess the complementarity of snow cover information derived from passive microwave imagery
(SMMR and SSM/I data) and those resulting from the visual interpretation of satellite images
gathered in the visible and near-IR portion of the electromagnetic spectrum (NOAA–NESDIS
weekly snow charts). Snow water equivalent variability observed during the period of peak snow
accumulation, snowmelt dates, annual snow cover duration, and dates of snow cover appearance and
disappearance are as many parameters whose fluctuations should be monitored to develop a valuable
index of climatic variability and climate change (Goodison and Walker, 1993; Barry et al., 1995).
We precisely focused here on estimating the dates of snow cover appearance and disappearance.
STUDY AREA
The study area is located in north Manitoba (Canada), on the western side of Hudson Bay,
between latitudes 56°N and 59°N, in an area dominated by wetlands and where terrain elevation
does not exceed 200 m (Hudson Bay Lowlands) (Figure 1).
One particular characteristic of this area is that it encompasses both the southern limit of the
continuous permafrost zone and northern limit of treeline and the continuous forest (Rouse, 1991).
The study area falls within the Subarctic ecoclimatic zone, a very contrasted geographical
environment that successively juxtaposes low density boreal forest, boreal forest–tundra transition
zone then shrubby tundra. Approximately 54.47% of the surface area is covered by evergreen
needleleaf forest, 26.08% by open areas (shrub/lichens, treeless and burns) and 10.30% by a
transition treed shrubland where the tree cover represents less than 10% of the vegetation
community (Figure1). These values were derived using the land cover map of Canada produced
from the NOAA–AVHRR imagery (Cihlar and Beaubien, 1998). This complex mosaic of cover
types is characterised by extremely different degrees of roughness and aerodynamic properties
influencing snow accumulation patterns, blowing snow and, overall, the satellite remotely sensed
signals.
Figure 1: Location of the study area and land cover.
17
MATERIAL AND METHODS
SMMR and SSM/I passive microwave data time series: data description, spectral signature
of snow and Principal Component Analysis
Passive microwave data from the Scanning Multichannel Microwave Radiometer (SMMR) and
the Special Sensor Microwave Imager (SSM/I), projected on an Equal-Area Scalable Earth Grid
(EASE-Grid) and corresponding to the 18 (19) and 37 GHz brightness temperatures, respectively,
recorded twice daily at the time of the Nimbus-7 and DMSP satellites repeated ascending and
descending overpasses (Nimbus-7 SMMR and DMSP SSM/I Pathfinder Daily EASE-Grid
Brightness Temperatures), were used in the current study (Armstrong et al., 1994 ; Knowles et al.,
1999). The SMMR and SSM/I radiometers provide, on a daily basis, digital charts of brightness
temperatures with a resampled spatial resolution of 25 km for the 18 (19) and 37 GHz channels.
Specific information on the EASE-Grid data product can be found in Armstrong and Brodzik
(1995).
A dual-frequency algorithm derived from the spectral gradient method of Künzi et al. (1982)
was developed and applied to the SMMR and SSM/I data. This algorithm was used twice: 1) for
computing the spectral difference between the daily brightness temperatures measured at 18 (19)
and 37 GHz, both in horizontal and vertical polarisation and 2) for averaging, on a five-day basis
(pentads), these spectral differences. Indeed, pentads are commonly applied to long time series of
passive microwave data, mainly in order to reduce the volume and computing time of the data to
process and, if required, to deal with the problem of orbital gaps recurrent below 56°N (Basist et
al., 1996) and occasional missing swath data (e.g. substantial SSM/I data gaps observed during the
1994–1995 period, especially for Alaska, Canadian Prairies and the surroundings areas).
The dual-frequency algorithm allows to determine the presence or absence of snow on the
ground and constitutes a valuable snow water equivalent index when applied to dry snow-covered
surfaces, in homogeneous and non-complex areas such as the Canadian Great Plains and the
Russian Steppes (Walker and Goodison, 1993). However, in the particularly heterogeneous
landscape encountered in the Subarctic ecoclimatic zone, passive microwave remote sensing of
snow cover comes up against the complexity of the interactions between microwave radiation and
the internal structure of snow cover, as well as the variability of the microwave signal with the
various cover types. This is the main reason why the experiment carried out by Boudreau and
Rouse (1994) in the vicinity of the northern village of Churchill (see figure 1), and which object
was to evaluate the potential of SSM/I passive microwave data to estimate snow physical
parameters, was not conclusive. Nevertheless, we demonstrated, in a previous study undertaken
around the same site (Pivot et al., 2002), that the SSM/I radiometer remains a very valuable tool
for monitoring the temporal evolution of snow cover in Subarctic environments. Thus, the
fluctuations of the multi-frequency algorithm described above could be regarded as a proxy of
snow cover conditions. Consequently, by interpreting these fluctuations, one is likely to obtain
accurate estimates of the dates of beginning, melting and disappearance of snow.
The passive microwave signature of dry snow strongly differs from the signature of snow-free
surfaces. For dry snow conditions, emissivities decrease with increasing frequency. Indeed, an
important microwave scattering commonly occurs above 30 GHz that jointly increases with the
amount of snow on the ground (Mätzler, 1987). Consequently, in the presence of dry snow, the
spectral difference is positive and increases gradually and concurrently with snow accumulation.
In opposition, for snow-free conditions, emissivities increase with increasing frequency (Mätzler,
1987). Consequently, when the ground is free of snow, the difference between the brightness
temperatures measured at frequencies 18 (19) and 37 GHz is negative. The microwave signature
of wet snow is similar to that of a snow-free surface and opposite to that of a dry snow cover;
emissivities increase with increasing frequency and reach, beyond 10 GHz, a maximum around 0.9
(Mätzler, 1987). This results in a generalised increase of brightness temperatures while the
difference between 18 (19) and 37 GHz frequencies tends to be considerably smaller. Therefore,
the multi-frequency algorithm remains positive during snowmelt, but tending towards values close
to 0 K.
18
A subset window (78750 km2) corresponding to the study area (Subarctic ecoclimatic zone for
Northern Manitoba) was extracted from the available SMMR and SSM/I Northern Hemisphere
images acquired over a 23-year period, from October 1978 to March 2001 (see the black box on
Figure 1). Sea-ice/water and coastal mixed pixels contaminated by the Hudson Bay were carefully
removed. A Principal Component Analysis (PCA), a well-known statistical technique to process
such voluminous time series, was performed to efficiently extract the snow cover information
contained in this extended image data set. PCA proceeds by a linear transformation of a series of
numeric variables in order to create a new series of orthogonal factors (principal components or
PCs) which, unlike the input variables, are uncorrelated between them. The PCs are ordered in
terms of the percentage of the explained variance contained in the original data set (Eastman and
Fulk, 1993).
In order to emphasise the temporal patterns contained in such time series of subset images, a
column-standardised principal component analysis with an orthogonal Varimax rotation was
performed. Indeed, column-standardisation was applied in order to obtain a variance of one for
each variable (i.e. spatial dimension), thus reducing the impact of the spatial variations induced by
the variety of landscapes and inversely highlighting the information related to the temporal
changes (Fung and LeDrew, 1987).
Extraction of the dates of snow cover appearance and disappearance from the NOAA–
NESDIS weekly Northern Hemisphere snow charts
The amount of solar radiation reflected over snow-covered surfaces considerably contrasts with
that of snow-free surfaces (Rango, 1993). As a result, visible-band/NIR spaceborne sensors
contributed early-on to the mapping snow cover extent. Since 1972, a series of meteorological
satellites was launched by the U.S. National Oceanic and Atmospheric Administration (NOAA),
that successively carried the VHRR (Very High Resolution Radiometer) and the AVHRR
(Advanced Very High Resolution Radiometer) sensors (Hall et Martinec,1985 ; Robinson, 1993).
The snow extent maps derived from these sensors and produced by the NOAA–NESDIS (NOAA
Environmental Satellite, Data and Information Service) were so accurate that they were soon
recognised as an operational and suitable product for hemispheric and regional-scale snow cover
studies. These charts are derived from the visual interpretation of the shortwave imagery by
trained analysts and are initially digitised on a weekly basis from a set of 89×89 cells projected
onto a polar stereographic grid. A cell is considered to be completely covered if it is interpreted to
be at least fifty percent snow-covered, otherwise it is considered to be snow free. A complete
description of the procedure can be found in Robinson (1993). Recently, the digital NOAA–
NESDIS Weekly Northern Hemisphere Snow Charts were regridded to the Northern Hemisphere
25-km EASE-Grid format (Armstrong and Brodzik, 2002), hence facilitating direct comparison
with other EASE-Grid products.
Thus, the matching subset window corresponding to that previously selected for the SMMR–
SSM/I EASE-Grid data was extracted over the same observation period. However, this EASEGrid subset window initially corresponded to four original cells with a poorer spatial resolution of
1×1 degree latitude/longitude. As a result, the effective subset area largely exceeded the
boundaries of the SMMR–SSM/I study area, especially in the southern part of the Subarctic zone,
where it encompasses a substantial part of the High-Boreal evergreen needleleaf forest.
Each year, the dates of snow cover appearance and disappearance were derived according to the
following principles: 1) the date of onset of snow corresponds to the first week where at least one
EASE-Grid cell is identified as snow-covered and 2) the date of snow disappearance matches the
week where all the cells of the subset area are concurrently interpreted as snow-free.
Conventional snow depth measurements and surface meteorological data
The dates of snow cover onset and disappearance derived from the NOAA–NESDIS weekly
snow charts and obtained from the SMMR–SSM/I passive microwave time series using PCA were
empirically compared to conventional snow cover and meteorological ground-based
measurements recorded at both the Churchill (58°44 N, 94°04 W) and Gillam (56°21 N, 94°42 W)
airport weather stations. These data are archived and distributed in digital format by the
19
Meteorological Service of Canada. Two major problems are related to traditional ground
measurements: the accuracy and the representativeness of these point data. As an example, the
Churchill site is located in the vicinity of the Hudson Bay, in an open area strongly exposed to
wind. The redistribution of snow by wind remains a key factor for explaining the spatial
variability of snow accumulation and the difficulty in estimating consistent and averaged snow
cover conditions using point measurements. Indeed, wind-effect is usually responsible for an
underestimation of snowfall catches by precipitation gauges; inversely, abnormal values can be
measured in the field during snow depth surveys, when snow has been significantly blown and
redistributed by wind. In addition, the Churchill weather station is located in a coastal area where
the snow cover conditions strongly differ from those prevailing in the close environment situated
farther inland. In fact, several vegetation types follow one another along a distance which does not
exceed 50 km and where one promptly passes from open forest, to partially treed transition zone and
then to tundra. This diversity definitely influences snow cover parameters and their evolution. At the
local scale, snow measurements made at the Churchill weather station are not representative of the
surroundings. Therefore, according to the heterogeneity of our study area on a regional scale (see
figure 1), conventional point measurements of snow cover should be considered with caution when
used as absolute values for validating the snow information derived from remote sensing techniques.
RESULTS AND DISCUSSION
PC2 scores of the PCA applied on 23 years of daily passive microwave data (SMMR–SSM/I
times series)
The sensitivity of passive microwave sensors (SMMR and SMM/I) data in identifying the
seasonal cycle and the interannual variability of snow cover in the Canadian Subarctic was
assessed by applying a PCA on the spectral differences computed as described previously, over a
23-year period (1978–2001). PC2 appears to be the most interesting component and explains
45.1% of the variance. The formation of this component is mainly attributed to the snow-covered
periods corresponding to spectral differences above 0K. Thus, annual snow-cover duration can be
estimated through this component since the reversal of the scores sign closely agrees with the
dates of appearance and disappearance of snow (Figure 2). Positive scores are commonly
interpreted as snow-free days or pentads (spectral differences below 0K), while negative scores as
periods where snow is present. However, negative scores are also noticeable in the early and
middle part of the snow cover season. There, they are associated with sudden changes affecting
the spectral-difference values (close to 0K), usually due to an episodic snowmelt event.
Furthermore, PC2 also depicts the interannual variability of snow cover, the strength of the scores
closely matching snow depth fluctuations. Identical results have previously been obtained with the
SSM/I times series alone (Pivot et al., 2002). However, unlike the SSM/I scores, those from
SMMR do not reflect the interannual variability of snow depth recorded at the Churchill airport
weather station. As an example, winter 1982–1983 has been the snowiest of the 1980s, but its
scores are almost at the same intensity as the scores recorded in 1984–1985, when snow depth
measured at the Churchill airport was approximately 50% lower. Preliminary observations assume
that the estimation of the dates of snow cover onset and disappearance is apparently not affected
by this bias if the scores at horizontal polarisation are employed. Nevertheless, there is a
perceivable sizeable shift at the vertical polarisation between the SMMR and SSM/I mean scores
intensity. For the moment, we are not able to provide some explanations and this problem is
currently under investigation. However, its consequence on snow cover studies can be clearly
established.
Obviously, this bias is likely to affect the estimation of the dates of snow cover periods from
PC2 scores insofar as the principle of the reversal of the scores sign should be maintained.
Moreover, considering the superiority of the vertical polarisation for retrieving snow cover
parameters, as demonstrated by research groups in Finland (Hallikainen and Jolma, 1992) and in
20
Canada (Goodison and Walker, 1993), this apparent bias could strongly affect snow water
equivalent retrieval algorithms which include vertical polarisation.
Figure 2: Interpretation of the PCA second axe based on snow depth measurements at the Churchill airport
weather station and comparison between the SMMR and SSM/I PC2
daily scores, for both horizontal and vertical polarisation.
Seasonal and interannual evolution of snow cover as detected by SMMR–SSM/I (horizontal
polarisation) and NOAA–AHVRR, and observed at the Churchill and Gillam airports
weather stations
The dates characterising the first snow accumulation and snow cover disappearance, obtained by
applying a PCA to SMMR and SSM/I daily and 5-day averaged spectral differences in horizontal
polarisation (for preventing the effects related to the SMMR data bias identified above), were
compared to both snow cover information derived from optical satellite imagery (NOAA–NESDIS
weekly snow charts) and conventional snow cover and meteorological measurements recorded at
the Churchill and Gillam airport weather stations.
In Figure 3, it appears that the passive microwave radiometers have some difficulties in
estimating accurately the date of the first snow on the ground when pentads are used instead of
daily observations, especially under warm and dry winter conditions, such as those that occurred
during fall 1992–1993 (Figure 3a) and fall 1998–1999 (Figure 3b). Indeed, snow fall recorded
during October 1992 and 1998 were 10 to 12 cm lower than the 1961–1990 normal, and snow
depth never exceeded 10 cm throughout several weeks. The difference between the dates of snow
21
cover onset, derived from daily data or pentads by the way of PCA, is considerable and exceeds 4
weeks (see shadow boxes in Figures 3a and 3b).
Figure 3: Comparison between multi-source data acquired for two warm and dry winters (a – 1992–1993 and
b – 1998–1999) illustrating the loss of information related to the use of pentad
instead of daily passive microwave data in PCA.
This temporal shift can be explained as follows. Microwaves naturally penetrate shallow dry
snow covers and thus are practically not scattered by the snow particles. At that time, the presence
of snow is difficult to detect as the signal recorded by the spacecraft sensors mainly comes from
the underlying ground that radiometrically appears “warmer” than dry snow (Bernier, 1987;
Rango, 1993). Therefore, in the dry and shallow snow cover situation prevailing in October 1992
and 1998, the passive microwave signature of each cell contained in our study area was almost
identical to the signature of snow-free surfaces so that the spectral differences values closely
reached 0K. Furthermore, during the early part of the 1992–1993 snow-cover period, strong
fluctuations of air temperature were observed so that the diurnal air temperature rose above 0°C on
several occasions (Figure 3a). The daily fluctuations of the PC2 scores, either positive else
negative, are associated with variations in air temperature and the state of snow cover (i.e.
succession of thawing–freezing events). By computing the mean spectral differences over pentads,
a significant part of the information related to the daily variability of the state of snow cover is
definitely smoothed and lost. For each pentad, the averaged spectral difference values thus closely
reach 0K and are inferred on the PCA output as continuous snow-free periods. For the same
reasons, this problem can also occur in spring, especially when snow decay is well underway and
snow cover remains thin and wet. Snow depth recorded as a trace over a more or less long period
and late snowfall succeeding the whole disappearance of winter snow cover (see the shadow box
on Figure 3b) can be misinterpreted as snow-free periods as well.
As they were more consistent, we only retained, in Figure 4, the dates of the first appearance of
snow and its definitive disappearance inferred from the daily PC2 scores, in order to compare
them with the dates derived from the NOAA–NESDIS weekly snow charts and snow information
obtained from ground weather stations. Note that for the Churchill and Gillam snow information,
we reported in Figure 4 both the dates of onset and ending of snow covers, established through
much of the winter season over a continuous period and the time period characterised by snow
traces or non-persistent snow accumulations.
The preliminary results show that SMMR and SMM/I data can efficiently compensate for the
deficiencies of visible-NIR imagery in estimating the dates of snow cover appearance or
disappearance, especially when cloud cover remains over the study area for long periods due to the
presence of large atmospheric disturbances which prevents the detection of snow cover for extended
periods. This characteristic situation is usually observed in autumn, during the first snow fall events
(e.g. 1987–1988, 1990–1991, 1993–1994 and 1994–1995). As an example, the estimation of the date
22
of snow cover appearance in October 1993 can be improved by 15 days when retaining the date
given by passive microwave data, which moreover perfectly agrees with ground observations.
Figure 4: Comparison between the dates of onset and disappearance of snow cover derived from the SMMR–
SMM/I data and NOAA–NESDIS weekly snow charts and observed
at the Churchill and Gillam airport weather stations.
The results inversely demonstrate the need to improve snow cover duration estimates derived
from passive microwave data by adding information from optical sensors, if one is interested in
identifying the occurrence of trace of snow on the ground (small amount of snow usually below 1
cm) and periods characterised by non-persistent snow accumulations during the transition seasons,
such as autumn and spring. The moisture, radiative, and heat fluxes between the atmosphere and
the ground can be strongly affected by small amounts of snow or intermittent snow-covered
periods so that it is highly recommended to accurately identify the occurrence of such events. As
and example, NOAA–NESDIS weekly snow charts appeared to be very suitable in assessing the
presence of a discontinuous snow cover during the 1982, 1989 and 1992 springs, at last improving
the snow cover duration estimations for these snow season periods.
Finally, two biased situations can be highlighted in Figure 4. The first contributes in strongly
underestimating the date of snow cover onset in autumn 1995 (around 8–10 weeks later) and was
clearly identified as a mathematical artefact generated by the PCA which origin is currently under
investigation. The second tends to overestimate by about 5–6 weeks the date of snow disappearance
during the summer 1985 and was probably produced by a misidentification of the presence of snow
on the ground on date of 07/19 that might have inadvertently been confused with that of clouds.
23
CONCLUSION
Since 1995, the NOAA–NESDIS supports the development and implementation of prototypes
for daily automated snow mapping systems based on a wide variety of satellite and surface
observations, such as the Interactive Multisensor snow and ice mapping System (IMS), which
should definitely replace the current weekly snow charts product (Ramsay, 1998). More recently,
Romanov et al. (2000) developed a similar interactive application enabling the computerised
extraction of snow information on a global scale using both the SSM/I passive microwave imagery
and optical and thermal sensors onboard Geostationary Operational Environmental Satellite
(GOES). The improvement of these applications is essential for an accurate and operational snow
cover monitoring. However, the success of such systems largely depends on the way the multisensor satellite data are precisely translated into snow information.
Using a Principal Component Analysis applied to SMMR and SSM/I time series (daily and
pentads) over a 23-year period (1978–2001), we were able to evaluate the sensitivity of passive
microwave data in describing the seasonal cycle and the interannual variability of snow cover
conditions in the Canadian Subarctic (northern Manitoba). The PC2 scores derived from the
passive microwave data set captured well the beginning and ending of the snow season, merely
when daily data were used as input. Indeed, we demonstrated that although pentads are commonly
used in the processing of passive microwave time series, there can be a loss of information as
opposed to daily data, especially under warm and dry winter conditions.
It has been clearly demonstrated that the Nimbus-7 SMMR Pathfinder Daily EASE-Grid
Brightness Temperatures are biased, especially in vertical polarisation. The long time series of
passive microwave data collected since the end of 70s provide an unprecedented opportunity for
monitoring the seasonal and the interannual variability of snow-cover parameters of relevance to
climate change in northern Canada. Consequently, this bias needs to be studied thoroughly in
further investigations because of its immediate repercussion on snow water equivalent retrieval
algorithms applied to SMMR data.
REFERENCES
Armstrong, R. L. and M. J. Brodzik. 1995. An earth-gridded SSM/I data set for cryospheric
studies and global change monitoring. Advances in Space Research, 10, 155–163.
Armstrong, R.L. and M.J. Brodzik. 2002. Northern Hemisphere EASE-Grid Weekly Snow Cover
and Sea Ice Extent Version 2. Boulder, CO, USA: National Snow and Ice Data Center. CD-ROM.
Armstrong, R.L., K.W. Knowles, M.J. Brodzik and M.A. Hardman. 1994 - [1987–2001]. DMSP
SSM/I Pathfinder daily EASE-Grid brightness temperatures. Boulder, CO: National Snow and Ice
Data Center. Digital media and CD-ROM.
Barry, R. G., J. –M. Fallot and R. L. Armstrong. 1995. Twentieth-century variability in snowcover conditions and approaches to detecting and monitoring changes : status and prospects.
Progress in Physical Geography, 19(4), 520–532.
Bernier, P. Y. 1987. Microwave remote sensing of snowpack properties : potential and
limitations. Nordic Hydrology, 18(1), 1–20.
Boudreau, L.D. and W.R. Rouse. 1994. Algorithm testing of passive microwave monitoring of
snow water equivalent in a tundra environment, Final report, Atmospheric Environment Service,
Environment Canada, 70 p.
Cihlar, J. and J. Beaubien. 1998. Land cover of Canada Version 1.1. Special Publication,
NBIOME Project. Produced by the Canada Centre for Remote Sensing and the Canadian Forest
Service, Natural Resource s Canada. Available on CD-ROM from the Canada Centre for Remote
Sensing, Ottawa, Ontario.
Eastman, J.R. and M. Fulk. 1993. Long sequence time series evaluation using standardized
principal components. Photogrammetric Engineering and Remote Sensing, 59(6), 991–996.
Fung, T. and E. Ledrew. 1987. Application of principal components analysis to change
detection. Photogrammetric Engineering and Remote Sensing, 53(12), 1649–1658.
24
Gloersen, P., D. J. Cavalieri, A. T. C. Chang, T. T. Wilheit, W. J. Campbell, O. M. Johannessen,
K. B. Katsaros, K. F. Kunzi, D. B. Ross, D. Staelin, E. P. L. Windsor, F. T. Barath, P. Gudmansen,
E. Langham, and R. Ramseier. 1984. A summary of results from the first Nimbus-7 SMMR
observations. Journal of Geophysical Research, 89(D4), 5335–5344.
Goodison, B. E. and A. E. Walker. 1993. Use of snow cover derived from satellite passive
microwave data as an indicator of climate change. Annals of Glaciology, 17, 137–142.
Hall, D. K. and J. Martinec. 1985. Remote sensing of ice and snow. Remote sensing
applications, Chapman et Hall, Grande Bretagne, 189 p.
Hallikainen M.T. and P.A. Jolma. 1992. Comparison of algorithms for retrieval of snow water
equivalent from Nimbus-7 SMMR data in Finland. IEEE. Transactions on Geoscience and
Remote Sensing, GE-30(1), 124–131.
Jin, Y.-Q. and N. Zhang. 1999. Correlation of the ERS and SSM/I observations over snowpack
and numerical simulation. International Journal of Remote Sensing, 20(15–16), 3009–3018.
Knowles, K.W., E.G. Njoku, R.L. Armstrong, and M.J. Brodzik. 1999 - [1978–1987]. Nimbus-7
SMMR Pathfinder daily EASE-Grid brightness temperatures. Boulder, CO: National Snow and
Ice Data Center. Digital media and CD-ROM.
Koskinen, J., S. Metsämäki, J. Grandell, S. Jänne, L. Matikainen and M. Hallikainen. 1999. Snow
monitoring using radar and optical satellite data. Remote Sensing of Environment, 69(1), 16–29.
Künzi, K. F., S. Patil and H. Rott. 1982. Snow-cover parameters retrieved from Nimbus-7
Scanning Multichannel Microwave Radiometer (SMMR) data. IEEE. Transactions on
Geoscience and Remote Sensing, 20(4), 452–467.
Mätzler, C. 1987. Applications of the interaction of microwaves with the natural snow cover.
Remote Sensing Reviews, 2, 259–392.
Pivot, F., C. Kergomard, and C.R. Duguay. 2002. On the use of passive microwave data to
monitor spatial and temporal variations of snowcover near Churchill, Manitoba. Annals of
Glaciology, 34, 58–64.
Ramsay, B. H. 1998. The interactive multisensor snow and ice mapping system. Hydrological
Processes, 12, 1537–1546.
Rango, A. 1993. II. Snow hydrology processes and remote sensing. Hydrological Processes,
7(2), 121–138.
Robinson, D. A. 1993. Hemispheric snow cover from satellites. Annals of Glaciology, 17, 367–
370.
Romanov, P., G. Gutman and I. Csiszar. 2000. Automated monitoring of snow cover over North
America with multispectral satellite data. Journal of Applied Meteorology, 39(11), 1866–1880.
Rouse, W.R. 1991. Impacts of Hudson Bay on the terrestrial climate of the Hudson Bay Lowlands.
Circumpolar Ecosystems in Winter: a Symposium and Workshop, Churchill, Manitoba,
Canada, 24–30.
Standley, A. P. and E. C. Barrett. 1999. The use of coincident DMSP SSM/I and OLS satellite
data to improve snow cover detection and discrimination. International Journal of Remote
Sensing, 20(2), 285–305.
Walker, A.E. and B.E. Goodison. 1993. Discrimination of a wet snow cover using passive
microwave satellite data. Annals of Glaciology, 17, 307–311.
25